基于长短期记忆神经网络的冷水机组能耗预测与用能异常分析OA

Based on long short-term memory(LSTM)neural networks,a collaborative optimization method for energy consumption forecasting and anomaly analysis is proposed.By collecting real-time operational data from a chiller unit in an electronics factory,and utilizing the Random Forest algorithm to select 12 key features strongly correlated with energy consumption,an LSTM forecasting model is developed.Dynamic thresholds are then set by comparing the predicted values with the actual values to identify potential energy anomalies.The experimental results show that the model's mean absolute error on the test set is 0.91,the root mean square error is 1.30,and the coefficient of determination reaches 0.97.The prediction accuracy of the model exceeds that of recurrent neural network and gated recurrent unit models.With the upper threshold set at 4.73 and the lower threshold at-1.07,532 abnormal data points are identified.The research indicates that this method,by integrating energy consumption prediction with abnormal energy usage analysis,can effectively identify abnormal energy consumption,providing a new technical path for the intelligent operation and maintenance and energy conservation management of chiller units.